Skinning method and apparatus, computer device, and storage medium

Abstract

A skinning method includes: acquiring a 3D mesh for skinning of a skeletal structure of a 3D object, the 3D mesh including a plurality of vertices and the skeletal structure including a plurality of joints; respectively acquiring global joint weight information corresponding to the plurality of vertices, the global joint weight information of a vertex including associated joints of the vertices predicted based on the 3D mesh, and joint weight values of the associated joints; determining a target vertex to be locally predicted from the vertices of the 3D mesh based on the global joint weight information; performing a local prediction based on a vertex feature of the target vertex to obtain local joint weight information of the target vertex; and optimizing global joint weight information of the target vertex based on the local joint weight information for skinning of the 3D object using the optimized joint weight information.

What is claimed is: 
     
       1. A skinning method executed by a computer device, the method comprising:
 acquiring a three-dimensional (3D) mesh for skinning a skeletal structure of a 3D object, the 3D mesh comprising a plurality of vertices and the skeletal structure comprising a plurality of joints; 
 respectively acquiring global joint weight information corresponding to the plurality of vertices, the global joint weight information corresponding to a vertex comprising associated joints of the vertex predicted based on the whole 3D mesh and joint weight values of the associated joints; 
 determining a target vertex to be locally predicted from the vertices of the 3D mesh based on the global joint weight information; 
 performing a local prediction based on a vertex feature of the target vertex to obtain local joint weight information of the target vertex, the local joint weight information comprising a local associated joint of the target vertex and a local joint weight value corresponding to the local associated joint; the local joint weight value indicating a degree of association between the local associated joint and the target vertex, and the local associated joint of the target vertex being an associated joint of the target vertex predicted based on a local mesh region in the 3D mesh; and 
 optimizing global joint weight information of the target vertex based on the local joint weight information to obtain optimized joint weight information, and generating skinned mesh of the 3D object using the optimized joint weight information. 
 
     
     
       2. The method according to  claim 1 , wherein the determining a target vertex to be locally predicted from the vertices of the 3D mesh based on the global joint weight information comprises:
 determining a vertex corresponding to abnormal global joint weight information to obtain an abnormal vertex; and 
 determining a target vertex to be locally predicted from the vertices of the 3D mesh based on the abnormal vertex, the target vertex comprising the abnormal vertex. 
 
     
     
       3. The method according to  claim 2 , wherein the 3D mesh comprises a plurality of mesh regions, and the determining a target vertex to be locally predicted from the vertices of the 3D mesh based on the abnormal vertex comprises:
 determining a mesh region to which the abnormal vertex belongs from the plurality of mesh regions of the 3D mesh to obtain an abnormal mesh region; and 
 determining the target vertex to be locally predicted based on vertices in the abnormal mesh region. 
 
     
     
       4. The method according to  claim 1 , wherein the performing a local prediction based on a vertex feature of the target vertex to obtain local joint weight information of the target vertex comprises:
 determining, for the target vertex, an adjacent joint having an adjacent relationship with the target vertex from the joints of the skeletal structure; 
 performing feature fusion based on a joint feature of the adjacent joint and the vertex feature of the target vertex to obtain a first local prediction reference feature corresponding to the target vertex; and 
 performing a local prediction based on the first local prediction reference feature to obtain the local joint weight information of the target vertex. 
 
     
     
       5. The method according to  claim 4 , wherein the determining an adjacent joint having an adjacent relationship with the target vertex from the joints of the skeletal structure comprises:
 determining a mesh region to which the target vertex belongs in the 3D mesh to obtain a target mesh region; 
 acquiring a joint corresponding to the target mesh region from the joints of the skeletal structure to obtain at least one candidate joint; 
 calculating a distance between the target vertex and the at least one candidate joint; and 
 selecting the adjacent joint having an adjacent relationship with the target vertex from the at least one candidate joint based on the distance. 
 
     
     
       6. The method according to  claim 4 , wherein the performing feature fusion based on a joint feature of the adjacent joint and the vertex feature of the target vertex to obtain a first local prediction reference feature corresponding to the target vertex comprises:
 acquiring a vertex feature corresponding to a neighbor vertex of the target vertex as a neighbor vertex feature; and 
 performing feature fusion based on the vertex feature of the target vertex, the joint feature of the adjacent joint, and the neighbor vertex feature to obtain the first local prediction reference feature corresponding to the target vertex. 
 
     
     
       7. The method according to  claim 6 , wherein the performing feature fusion based on the vertex feature of the target vertex, the joint feature of the adjacent joint, and the neighbor vertex feature to obtain the first local prediction reference feature corresponding to the target vertex comprises:
 determining a vertex feature difference between the neighbor vertex feature of the neighbor vertex and the vertex feature of the target vertex; 
 concatenating the vertex feature difference and the vertex feature of the target vertex to obtain a vertex concatenation feature of the target vertex; and 
 performing feature fusion based on the vertex concatenation feature and the joint feature of the adjacent joint to obtain the first local prediction reference feature corresponding to the target vertex. 
 
     
     
       8. The method according to  claim 4 , wherein the performing a local prediction based on the first local prediction reference feature to obtain the local joint weight information of the target vertex comprises:
 acquiring, for an adjacent joint of the target vertex, a target vertex having an adjacent relationship with the adjacent joint from the target vertex to be locally predicted, and obtaining an adjacent vertex corresponding to the adjacent joint; 
 performing feature fusion based on the joint feature of the adjacent joint and a vertex feature of the adjacent vertex to obtain a joint fusion feature corresponding to the adjacent joint; 
 performing feature fusion based on the joint fusion feature of the adjacent joint and the first local prediction reference feature to obtain a second local prediction reference feature corresponding to the target vertex; and 
 performing a local prediction based on the second local prediction reference feature to obtain the local joint weight information of the target vertex. 
 
     
     
       9. The method according to  claim 8 , wherein the performing feature fusion based on the joint feature of the adjacent joint and a vertex feature of the adjacent vertex to obtain a joint fusion feature corresponding to the adjacent joint comprises:
 performing a statistical operation on the vertex feature of the adjacent vertex to obtain a vertex statistical feature; and 
 performing feature fusion based on the joint feature of the adjacent joint and the vertex statistical feature to obtain the joint fusion feature corresponding to the adjacent joint. 
 
     
     
       10. The method according to  claim 9 , wherein the performing feature fusion based on the joint feature of the adjacent joint and the vertex statistical feature to obtain the joint fusion feature corresponding to the adjacent joint comprises:
 acquiring a joint feature of a neighbor joint corresponding to the adjacent joint as a neighbor joint feature; 
 performing feature fusion on the joint feature of the adjacent joint and the neighbor joint feature to obtain a preliminary fusion feature corresponding to the adjacent joint; and 
 performing feature fusion on the preliminary fusion feature corresponding to the adjacent joint and the vertex statistical feature to obtain the joint fusion feature corresponding to the adjacent joint. 
 
     
     
       11. The method according to  claim 8 , wherein the performing feature fusion based on the joint fusion feature of the adjacent joint and the first local prediction reference feature to obtain a second local prediction reference feature corresponding to the target vertex comprises:
 determining a neighbor vertex of the target vertex, and acquiring a first local prediction reference feature corresponding to the neighbor vertex; 
 performing feature fusion based on the first local prediction reference feature of the target vertex and the first local prediction reference feature of the neighbor vertex to obtain a local prediction fusion feature corresponding to the target vertex; and 
 performing feature fusion on the local prediction fusion feature and the joint fusion feature of the adjacent joint to obtain the second local prediction reference feature corresponding to the target vertex. 
 
     
     
       12. The method according to  claim 8 , wherein the performing a local prediction based on the second local prediction reference feature to obtain the local joint weight information of the target vertex comprises:
 determining a neighbor vertex of the target vertex, and acquiring a second local prediction reference feature corresponding to the neighbor vertex; 
 performing feature fusion on the second local prediction reference feature corresponding to the target vertex and the second local prediction reference feature of the neighbor vertex to obtain a target local prediction reference feature corresponding to the target vertex; and 
 performing a local prediction based on the target local prediction reference feature to obtain the local joint weight information of the target vertex. 
 
     
     
       13. The method according to  claim 12 , wherein the performing feature fusion on the second local prediction reference feature corresponding to the target vertex and the second local prediction reference feature of the neighbor vertex to obtain a target local prediction reference feature corresponding to the target vertex comprises:
 calculating differences between the second local prediction reference feature of the neighbor vertex and the second local prediction reference feature of the target vertex to obtain a second reference feature difference; 
 performing feature concatenation on the second reference feature difference and the second local prediction reference feature of the target vertex to obtain a second reference concatenation feature; and 
 obtaining the target local prediction reference feature corresponding to the target vertex based on the second reference concatenation feature. 
 
     
     
       14. A skinning apparatus, comprising:
 one or more processors and a memory storing computer-readable instructions, the computer-readable instructions, when executed by the one or more processors, causing the one or more processors to implement: 
 acquiring a three-dimensional (3D) mesh for skinning a skeletal structure of a 3D object, the 3D mesh comprising a plurality of vertices and the skeletal structure comprising a plurality of joints; 
 respectively acquiring global joint weight information corresponding to the plurality of vertices, the global joint weight information corresponding to a vertex comprising associated joints of the vertex predicted based on the whole 3D mesh and joint weight values of the associated joints; 
 determining a target vertex to be locally predicted from the vertices of the 3D mesh based on the global joint weight information; 
 performing a local prediction based on a vertex feature of the target vertex to obtain local joint weight information of the target vertex, the local joint weight information comprising a local associated joint of the target vertex and a local joint weight value corresponding to the local associated joint; the local joint weight value indicating a degree of association between the local associated joint and the target vertex, and the local associated joint of the target vertex being an associated joint of the target vertex predicted based on a local mesh region in the 3D mesh; and 
 optimizing global joint weight information of the target vertex based on the local joint weight information to obtain optimized joint weight information, and generating skinned mesh of the 3D object using the optimized joint weight information. 
 
     
     
       15. The apparatus according to  claim 14 , wherein the determining a target vertex to be locally predicted from the vertices of the 3D mesh based on the global joint weight information comprises:
 determining a vertex corresponding to abnormal global joint weight information to obtain an abnormal vertex; and 
 determining a target vertex to be locally predicted from the vertices of the 3D mesh based on the abnormal vertex, the target vertex comprising the abnormal vertex. 
 
     
     
       16. The apparatus according to  claim 15 , wherein the 3D mesh comprises a plurality of mesh regions, and the determining a target vertex to be locally predicted from the vertices of the 3D mesh based on the abnormal vertex comprises:
 determining a mesh region to which the abnormal vertex belongs from the plurality of mesh regions of the 3D mesh to obtain an abnormal mesh region; and 
 determining the target vertex to be locally predicted based on vertices in the abnormal mesh region. 
 
     
     
       17. The apparatus according to  claim 14 , wherein the performing a local prediction based on a vertex feature of the target vertex to obtain local joint weight information of the target vertex comprises:
 determining, for the target vertex, an adjacent joint having an adjacent relationship with the target vertex from the joints of the skeletal structure; 
 performing feature fusion based on a joint feature of the adjacent joint and the vertex feature of the target vertex to obtain a first local prediction reference feature corresponding to the target vertex; and 
 performing a local prediction based on the first local prediction reference feature to obtain the local joint weight information of the target vertex. 
 
     
     
       18. The apparatus according to  claim 17 , wherein the determining an adjacent joint having an adjacent relationship with the target vertex from the joints of the skeletal structure comprises:
 determining a mesh region to which the target vertex belongs in the 3D mesh to obtain a target mesh region; 
 acquiring a joint corresponding to the target mesh region from the joints of the skeletal structure to obtain at least one candidate joint; 
 calculating a distance between the target vertex and the at least one candidate joint; and 
 selecting the adjacent joint having an adjacent relationship with the target vertex from the at least one candidate joint based on the distance. 
 
     
     
       19. The apparatus according to  claim 17 , wherein the performing feature fusion based on a joint feature of the adjacent joint and the vertex feature of the target vertex to obtain a first local prediction reference feature corresponding to the target vertex comprises:
 acquiring a vertex feature corresponding to a neighbor vertex of the target vertex as a neighbor vertex feature; and 
 performing feature fusion based on the vertex feature of the target vertex, the joint feature of the adjacent joint, and the neighbor vertex feature to obtain the first local prediction reference feature corresponding to the target vertex. 
 
     
     
       20. One or more non-transitory readable storage media storing computer-readable instructions, the computer-readable instructions, when executed by one or more processors, causing the one or more processors to implement:
 acquiring a three-dimensional (3D) mesh for skinning a skeletal structure of a 3D object, the 3D mesh comprising a plurality of vertices and the skeletal structure comprising a plurality of joints; 
 respectively acquiring global joint weight information corresponding to the plurality of vertices, the global joint weight information corresponding to a vertex comprising associated joints of the vertex predicted based on the whole 3D mesh and joint weight values of the associated joints; 
 determining a target vertex to be locally predicted from the vertices of the 3D mesh based on the global joint weight information; 
 performing a local prediction based on a vertex feature of the target vertex to obtain local joint weight information of the target vertex, the local joint weight information comprising a local associated joint of the target vertex and a local joint weight value corresponding to the local associated joint; the local joint weight value indicating a degree of association between the local associated joint and the target vertex, and the local associated joint of the target vertex being an associated joint of the target vertex predicted based on a local mesh region in the 3D mesh; and 
 optimizing global joint weight information of the target vertex based on the local joint weight information to obtain optimized joint weight information, and generating skinned mesh of the 3D object using the optimized joint weight information.